Document 13378200

advertisement
Detection of In-band Wormholes Using Sequential Change Detection Algorithms
Shanshan Zheng, Tao Jiang, John. S. Baras
Background
Sequential Change Detection
Algorithms for the Leaf Nodes
Wormhole Attacks in Wireless Ad Hoc Networks:
 
  Malicious nodes 3 and 8 create the illusion that they are one-hop neighbors by
using a covert communication tunnel.
  Based on the tunneling scheme, wormholes can be classified as in-band
wormholes and out-band wormholes.
In-band wormhole
Out-band wormhole
Purported neighbors are connected
via multi-hop tunnels over existing
wireless medium, do not need
additional hardware, more likely to
be used by adversaries
Purported neighbors are connected
using an external communication
medium, (e.g., a wired link), may
need additional specialized
hardware
consume network capacity
add channel capacity to network
Countermeasures not depend on
attack mechanism
Countermeasures depend on attack
mechanism
Our work deals with in-band wormholes.
  Threats of the wormhole attack
  Undermines shortest path routing calculations
  Create artificial traffic choke points under control of the attacker
Framework of Our Detection
Scheme
  Motivation
A in-band wormhole attack leads an abrupt change in the transmission
delay along a path
formulate in-band wormhole detection as a
sequential change detection problem
  Detection Scheme
Leaf nodes: collect 3-hop
transmission delay, make
individual inferences, send them
to the cluster head
Cluster head: correlates
individual inferences it receives
to make final decision and
locate the wormhole
Two Sequential Change Detection algorithms are proposed to help the leaf node
make individual inferences about whether there is a wormhole or not
  Non-parametric Cumulative Sum (NP-CUSUM)
  Repeated Sequential Probability Ratio Test (R-SPRT)
  NP-CUSUM
•  Used when an attack model is not available (e.g., network topology
changes quickly, makes it difficult to estimate the distributions of the
path delay measurement)
•  Statistic, decision rule and stopping time for NP-CUSUM:
Hynet
Simulations
  Simulation setting:
  NS-2 simulator, OLSR based wireless ad hoc network
  50 nodes in a 1000x1000 square field
  2 attackers form an 8-hop wormhole
  Some results:
  Performance of an NP-CUSUM based detector and a R-SPRT based
detection
Statistic:
Decision rule:
Stopping time:
is the delay measurement obtained by the leaf node at time
n, h is a threshold and c is some pre-defined constant
  R-SPRT
•  Used when an attack model is available
•  Statistic, decision rule and stopping time for a single SPRT:
 
Performance of a R-SPRT based detector using improper training
set
Statistic:
Decision rule:
Stopping time:
•  In R-SPRT, the single SPRT is restarted whenever a ‘0’ decision is made.
This setup enables continuous monitoring of wormhole detection
Correlation Algorithm for the
Cluster Head
  For the cluster head, locating a wormhole requires at least two
anomalous observations with a common intersecting link but disjoint
end nodes.
A
B
E
D
A-B-C-D
?
F-C-B-E
•  Better performance of R-SPRT than NP-CUSUM comes at the
cost of more attack information and higher computational
complexity
•  Using improper training data for R-SPRT can seriously degrade
its performance
Acknowledgement
This work is sponsored by the U.S. Army Research Laboratory under
cooperative agreement DAAD19-01-2-0011.
C
F
Download